#951 Fix warning: unused parameter
Metadata
| Property | Value |
|---|---|
| Reported by | Elias Pipping (elias.pipping@fu-berlin.de) |
| Reported at | Sep 14, 2011 15:01 |
| Type | Bug Report |
| Version | 2.1 |
| Operating System | Unspecified / All |
| Last edited by | Oliver Sander (oliver.sander@tu-dresden.de) |
| Last edited at | Sep 19, 2011 09:33 |
| Closed by | Oliver Sander (oliver.sander@tu-dresden.de) |
| Closed at | Sep 19, 2011 09:33 |
| Closed in version | Unknown |
| Resolution | Won't fix |
| Comment | Sorry. |
Description
Patches against the 2.1 branch.
Attachments
Activity
Not really related but also a patch against the 2.1 branch of dune-common (that I probably shouldn't open a separate bug for)
Attachments
I do not like the two patches.By omitting the parameters there is now way for documenting them in doxygen.
Maybe we could resort the patches to use the third approach of http://www.agavemountain.com/2007/09/dealing-with-c-unused-parameter.html ?
-
-Wunused-parameter is implied by "-Wall -Wunused" and "-Wextra" but not plain "-Wall" (which is the autotools default).
The #pragma unused invocation is not too ugly and has the advantage of being ignored by any compiler that does not support it.
There is also attribute((unused)) but that's not only long and ugly, but also only supported by gcc I think.
One could also say that if doxygen is used with code that does not declare functions separately, unused variables cannot be avoided in a portable way, so that warnings about such should just be ignored/disabled.
-
One could also say that if doxygen is used with code that does not declare functions separately, unused variables cannot be avoided in a portable way, so that warnings about such should just be ignored/disabled.
In other words (for gcc): Whoever uses -Wunused (be that directly or through -Wextra) should pass -Wno-unused-parameter as well.
I agree with Markus, that we want to document the parameters. I see two possible ways:
- use the void cast ... kind of ugly, but it works ... might be made a bit more readable with the "#define UNUSED_ARGUMENT(x) (void)x", as suggested on the webpage Markus mentioned.
- use something like void foo(int UNUSED_ARG(bar)) where UNUSED_ARG might be defined to different things, when compiling, or when using doxygen. For doxygen we would define something like "#define UNUSED_ARG(x) x", whereas the normal dune code would contain a definition like "#define UNUSED_ARGUMENT(x)" or "#define UNUSED_ARG(x) UNUSED_ ## x attribute((unused))" if attribute unused is supported.
-
Then a function like the following
virtual void foo(int i, double x, double y, Interval& d, int j) const { bar(i, d, j); return; }
would turn into...
With (1):
virtual void foo(int i, double x, double y, Interval& d, int j) const { (void)x; (void)y; bar(i, d, j); return; }
or with a #define:
virtual void foo(int i, double x, double y, Interval& d, int j) const { UNUSED_ARGUMENT(x); UNUSED_ARGUMENT(y); bar(i, d, j); return; }
And with (2):
virtual void foo(int i, double UNUSED_ARG(x), double UNUSED_ARG(y), Interval& d, int j) const { bar(i, d, j); return; }
Is that really desirable?
The last version adds the least amount of noise imho.
